CAREER: Enabling Perception-Driven Optimization for Online Videos

该奖项全部或部分由《2021年美国救援计划法案》（公法117-2）资助。今天，视频流媒体不再仅仅是娱乐，而是我们日常生活中必不可少的一部分。例如，高分辨率的实时视频用于日常（远程）学习，并且不断分析来自路边传感器的视频流，以使我们的城市和道路更安全。这些变化导致对网络带宽的更高需求，以达到期望的体验质量（QoE）。这是具有挑战性的，因为为了满足所有这些需求，今天的视频传输系统必须要么降低视频质量，要么升级网络基础设施（这可能是缓慢和昂贵的）。该项目开发了感知驱动优化（PDO），这是一种新的范例，可以在不使用更多带宽的情况下改善当今在线视频的用户体验。关键的见解是，就视频质量如何影响QoE而言，视频、人类用户和视频分析模型之间存在显著的异构性，这很难被当今的离线QoE模型捕获。PDO采用数据驱动的方法，通过利用当今视频交付系统可用的大量视频会话，实现在线QoE建模的自动化。本研究涉及三个重点：(a)对于直播视频服务，如何在直播内容中以最少的在线会话数量自动化QoE建模；(b)对于点播视频，如何在不影响用户体验的情况下构建每个用户的QoE模型；(c)对于视频分析服务，如何分析视频质量对视频分析模式的影响。PDO还将在线QoE建模与当今的视频传输系统集成在一起，以更好地适应网络条件的变化。该项目与在线视频服务和边缘视频分析的行业合作伙伴合作，以及部署PDO和改善在线视频QoE的本地倡议，特别是针对遭受质量问题的用户，并将边缘分析扩展到更多传感器视频。它还创造了将系统/网络教育与互联网的日常使用联系起来的新方法，例如可视化网络性能如何影响用户感知的视频质量和基于视频的智能应用程序的新教育工具。该项目还通过针对弱势群体的领导力联盟（Leadership Alliance）与学生接触，开发的教育工具将用于针对高中女生的compileHer项目。作为这个项目的一部分，软件和研究成果在一个公共网站上发布：https://people.cs.uchicago.edu/~junchenj/perception_driven_optimization。该网站定期维护，包括发布的数据、源代码和复制说明。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This award is funded in whole or in part under the American Rescue Plan Act of 2021 (Public Law 117-2).Today, video streaming is no longer just about entertainment but essential to our daily life. For example, high-resolution live videos are used for daily (remote) learning, and videos streamed from road-side sensors are constantly analyzed to make our cities and roads safer. These changes lead to a higher need for network bandwidth to reach desired Quality of Experience (QoE). This is challenging, because to serve all these demands, today’s video delivery systems must either lower video quality or upgrade the network infrastructure (which can be slow and expensive). This project develops Perception-Driven Optimization (PDO), a new paradigm that improves user experience for today’s online videos without using more bandwidth. The key insight is that in terms of how video quality impacts QoE, significant heterogeneities exist across videos, human users, and video analytic models, which are hard to capture by today’s offline QoE models. PDO takes a data-driven approach to automate online QoE modeling, by leveraging the large number of video sessions available to today’s video delivery systems. This research entails three thrusts: (a) for live video services, how to automate QoE modeling in live content with a minimal number of online sessions; (b) for on-demand videos, how to build per-user QoE models without impacting user experience; and (c) for video-analytics services, how to profile video quality’s impact on video-analytics mode. PDO also integrates online QoE modeling with today’s video-delivery systems to better adapt to changes in network conditions.This project works with industry partners in online video services and edge video analytics, and local initiatives to deploy PDO and improve online video QoE, especially for users who suffer quality issues, and scale edge analytics to more sensor videos. It also creates new ways to tie system/networking education with everyday use of the Internet, such as new educational tools that visualize how network performance affects user-perceived video quality and video-based intelligent applications. The project also engages with students through Leadership Alliance which targets underrepresented populations, and the developed educational tools will be used for compileHer, a program that targets high school female students.The software and research artifacts implemented as part of this project are released on a public website: https://people.cs.uchicago.edu/~junchenj/perception_driven_optimization. The site is regularly maintained and includes released data, source code, and reproduction instructions.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.